Title :
Fruitless search in differential evolution
Author :
Engelbrecht, Andries
Author_Institution :
Dept. of Comput. Sci., Univ. of Pretoria, Pretoria, South Africa
Abstract :
Recent research has shown for different particle swarm optimization algorithms that unconstrained particles exhibit roaming behavior in that particles leave the boundaries of the search space very early during the search [1], [2]. This results in fruitless search of infeasible space, and will result in particles finding infeasible solutions if better solutions exist outside of the boundaries of the problem being optimized. This study evaluates the extent to which fruitless search due to roaming behavior occurs within differential evolution. Empirical results are provided that unconstrained individuals of a DE/rand/1/bin strategy also leave the search bounds, with the severity of fruitless wandering increasing with an increase in the scale parameter. Empirical evidence is also provided to show that infeasible solutions that violate boundary constraints will be found should better solutions exist outside of the defined boundaries.
Keywords :
evolutionary computation; particle swarm optimisation; search problems; differential evolution; empirical analysis; fruitless wandering; infeasible solutions; infeasible space fruitless search; particle swarm optimization algorithms; scale parameter; search bounds; search space boundary constraints; unconstrained DE/rand/1/bin strategy; unconstrained particle roaming behavior; Benchmark testing; Convergence; Optimization; Search problems; Sociology; Statistics; Vectors; Differential evolution; boundary constraints; roaming behavior;
Conference_Titel :
Differential Evolution (SDE), 2013 IEEE Symposium on
Conference_Location :
Singapore
DOI :
10.1109/SDE.2013.6601436